hank childs, university of oregon

Hank Childs, University of OregonApril 10th, 2015

CIS 330: _ _ _ _ ______ _ _____ / / / /___ (_) __ ____ _____ ____/ / / ____/ _/_/ ____/__ __ / / / / __ \/ / |/_/ / __ `/ __ \/ __ / / / _/_// / __/ /___/ /_ / /_/ / / / / /> < / /_/ / / / / /_/ / / /____/_/ / /__/_ __/_ __/ \____/_/ /_/_/_/|_| \__,_/_/ /_/\__,_/ \____/_/ \____//_/ /_/

Lecture 4:Memory

Announcements

• Matt’s OH: Mon 12-1, Tues 12-2• Hank’s OH: Weds 3:45-4:45, Fri 12:30-1:30

Outline

• Review• Intro to Memory Segments• Memory Allocation Basics• Project 2A• Advanced Memory Concepts / Memory Errors

Outline

Build: The Actors

• File types– Source code– Object code– Executable code

• Programs– Compiler– Linker

Source code(e.g., C code)

Compiler Object code Linker Executable code

Compilers, Object Code, and Linkers

• Compilers transform source code to object code– Confusing: most compilers also secretly have access

to linkers and apply the linker for you.• Object code: statements in machine code – not executable– intended to be part of a program

• Linker: turns object code into executable programs

GNU Compilers

• GNU compilers: open source– gcc: GNU compiler for C– g++: GNU compiler for C++

gcc flag “-c”: make object code, but don’t link

• gcc –c file1.c– makes object code file “file1.o”

Multi-file development: examplecat is a Unix commandthat prints the contentsof a file

$? is a shell construct thathas the return value of thelast executed program

Multi-file development: example

Linker order matters for some linkers (not Macs).

Some linkers need the .o with “main” first and then extract

the symbols they need as they go.

Other linkers make multiple passes.

Libraries

• Library: collection of “implementations” (functions!) with a well defined interface

• Interface comes through “header” files.• In C, header files contain functions and

variables.– Accessed through “#include <file.h>”

Making a static library

Note the ‘#’ is the comment character

(should have called this libmultiplier.a)

Typical library installations

• Convention– Header files are placed in “include” directory– Library files are placed in “lib” directory

• Many standard libraries are installed in /usr– /usr/include– /usr/lib

• Compilers automatically look in /usr/include and /usr/lib (and other places)

Installing the library

(fixing my mistake)

“mv”: unix command for renaming a file

Example: compiling with a library

• gcc support for libraries– “-I”: path to headers for library– “-L”: path to library location– “-lname”: link in library libname

Makefiles

• There is a Unix command called “make”• make takes an input file called a “Makefile”• A Makefile allows you to specify rules– “if timestamp of A, B, or C is newer than D, then

carry out this action” (to make a new version of D)• make’s functionality is broader than just

compiling things, but it is mostly used for computation

Basic idea: all details for compilation are captured in a configuration file … you just invoke “make” from a shell

Makefiles

• Reasons Makefiles are great:– Difficult to type all the compilation commands at a

prompt– Typical develop cycle requires frequent

compilation– When sharing code, an expert developer can

encapsulate the details of the compilation, and a new developer doesn’t need to know the details … just “make”

Makefile syntax

• Makefiles are set up as a series of rules• Rules have the format:

target: dependencies[tab] system command

Target: what to build. (a name you give to make.)Dependencies: what it depends on (files in the filesystem or other rules)System command: gcc …

Makefile example: multiplier lib

Configuration management tools

• Problem:– Unix platforms vary• Where is libX installed?• Is OpenGL supported?

• Idea:– Write problem that answers these questions, then

adapts build system• Example: put “-L/path/to/libX –lX” in the link line• Other fixes as well

Two popular configuration management tools

• Autoconf– Unix-based– Game plan:

• You write scripts to test availability on system• Generates Makefiles based on results

• Cmake– Unix and Windows– Game plan:

• You write .cmake files that test for package locations• Generates Makefiles based on results

CMake has been gaining momentum in recent years, because it is one of the best solutions for cross-platform support.

Unix command: tar

• Problem: you have many files and you want to…– move them to another machine– give a copy to a friend– etc.

• Tar: take many files and make one file– Originally so one file can be written to tape drive

• Serves same purpose as “.zip” files.

Unix command: tar

• tar cvf 330.tar file1 file2 file3• scp 330.tar @ix:~• ssh ix• tar xvf 330.tar• ls

file1 file2 file

Project 1C

Outline

Memory Segments

• Von Neumann architecture: one memory space, for both instructions and data

• so break memory into “segments”– … creates boundaries to prevent confusion

• 4 segments:– Code segment– Data segment– Stack segment– Heap segment

Code Segment

• Contains assembly code instructions• Also called text segment• This segment is modify-able, but that’s a bad

idea– “Self-modifying code”• Typically ends in a bad state very quickly.

Data Segment

• Contains data not associated with heap or stack– global variables– statics (to be discussed later)– character strings you’ve compiled in

char *str = “hello world\n”

Stack: data structure for collection

• A stack contains things• It has only two methods: push and pop– Push puts something onto the stack– Pop returns the most recently pushed item (and

removes that item from the stack)• LIFO: last in, first out

Imagine a stack of trays.You can place on top (push).

Or take one off the top (pop).

• Stack: memory set aside as scratch space for program execution

• When a function has local variables, it uses this memory.– When you exit the function, the memory is lost

• The stack grows as you enter functions, and shrinks as you exit functions.– This can be done on a per variable basis, but the

compiler typically does a grouping.• Some exceptions (discussed later)

• Don’t have to manage memory: allocated and freed automatically

• Heap (data structure): tree-based data structure

• Heap (memory): area of computer memory that requires explicit management (malloc, free).

• Memory from the heap is accessible any time, by any function.– Contrasts with the stack

Memory Segments

Source: http://www.cs.uwm.edu/classes/cs315/Bacon/

Outline

C: must manage your own memory

• This is a big change from other programs• You keep track of memory– Allocation– How much there is / indexing memory– Deallocation

malloc

• malloc: command for allocating memory

Allocation / Deallocation Example

Automatic allocation on thestack. (Deallocation occurs when out of scope.)

Explicit allocation from the heap. (Deallocation occurs with “free” call.)

sizeof

• sizeof: gets size of type• Usually:– sizeof(int) == 4– sizeof(float) == 4– sizeof(double) == 8– sizeof(unsigned char) == 1– sizeof(char) == 1

• array of 10 ints malloc(10*sizeof(int))

Hexadecimal

• Binary: 2 values• Decimal: 10 values• Hexadecimal: 16 values– 0, 1, 2, 3, 4, 5, 6, 7, 8, 9, A, B, C, D, E, F

• 0x: prefix for hexadecimal• 0x10 = 16• 0x101 = 257

Memory Addresses

• Every location in memory has an address associated with it

• Locations in memory are represented in hexadecimal

0x7fff55bc

0x7fff55b8

Memory addresses descend in the stack, ascend in the heap.

Pointers• Pointers store locations in memory

• “&”: unary operator that gives the address of a variable.

int x;int *yp = &x;

Pointers• Pointers store locations in memory

printf prints pointers with “%p”

NULL pointer• NULL: defined by

compiler to be a location that is not valid.– Typically 0x00000000

• You can use NULL to initialize pointers, and also to check to see whether a pointer is set already.

IBM team I worked on used 0xDEADBEEF, not NULL

‘*’ operator• Let “ptr” be a pointer• Then “*ptr” returns value in the address that pts

points to. • * = “dereference operator”

Behavior of dereference

• When you dereference, you get the value at that moment.– Whatever happens afterwards won’t have effect.

Pointer Arithmetic• You can combine pointers and integers to get

new pointer locations

ptr + 1 ptr + sizeof(type)

Arrays

• Arrays: container that has multiple elements of identical type, all stored in contiguous memory

int A[10];

10 integers, stored in 40 consecutive bytes (assuming sizeof(int) == 4)

Arrays are just pointers. You can use arrays and pointers interchangeably.

[ ] operator

• [ ] is a way of dereferencing memory– Recall that ‘*’ is the dereference operator

• A[0] <= => *A• A[5] <= => *(A+5);

More array relationships

int A[10];int *B;

B=(A+5) A[5] = B[0]B=&(A[0]) B = AB=&(A[5]) B = A+5

Pointers to pointers

• Remember: pointer points to a location in memory– We’ve been considering

cases where locations in memory are arrays of integers

– But locations in memory could be pointer themselves

0x7fff55bc 0x7fff55b8

Simple pointers to pointers example

What’s the difference between these two programs?

Answer: X is on the heap on the left, and on the stack on the right.But they are both pointers-to-pointers.

What’s the difference between these two programs?

Answer: program on left makes one allocation for each pointer,program on right makes one allocation for whole program & each pointer points at locations within that allocation.

Call by reference / call by value

• Refers to how parameters are passed to a function.– Call by reference: send a reference (pointer) as a

function parameter• Side effects in that function affect the variable in the

calling function– Call by value: send the value of the variable as a

function parameter• Side effects in that function don’t affect the variable in

the calling function

Call by Reference

Call by Value

Outline

• Review• Intro to Memory Segments• Memory Allocation Basics• Project 2D• Advanced Memory Concepts / Memory Errors

Called Project 2D since I screwed up and printed a bunch of them

Outline

Stack vs Heap: Pros and ConsStack Heap

Allocation/Deallocation

Automatic Explicit

How stack memory is allocated into Stack Memory Segment

stack_varCstack_varD

stack_varCstack_varDstack_varAstack_varB

stack_varCstack_varD<info for how to get back to main>A (= 3)<Location for RV>

stack_varCstack_varD<info for how to get back to main>A (= 3)<Location for RV>stack_varA

Return copies into location specified by calling function

stack_varC = 6stack_varD = 3

This code is very problematic … why?

foo and bar are returning addresses that are on the stack … they could easily

be overwritten(and bar’s stack_varD

overwrites foo’s stack_varC in this

program)

Nested Scope

stack_varA

Nested Scope

stack_varAstack_varB

Nested Scope

stack_varA

You can create new scope within a function by adding

‘{‘ and ‘}’.

Automatic Explicit

Access Fast Slower

Memory pages associated with stack are almost always immediately

available.

Memory pages associated with heap may be located

anywhere ... may be caching effects

Automatic Explicit

Access Fast Slower

Variable scope Limited Unlimited

Variable scope: stack and heap

bar returned memory from heap

The calling function – i.e., the function that

calls bar – must understand this and take responsibility for calling

If it doesn’t, then this is a “memory leak”.

Memory leaksCode

stack_varA

It is OK that we are using the heap … that’s what it is there for

The problem is that we lost the references to the 49 allocations on heap

The heap’s memory manager will not be able to re-claim them … we have effectively limited the

memory available to the program.

Running out of memory (stack)Code

Freestack overflow: when the stack runs into the heap.There is no protection for stack overflows.

(Checking for it would require coordination with the heap’s memory manager on every function calls.)

Running out of memory (heap)Code

FreeIf the heap memory manager doesn’t have room to make an allocation, then malloc returns NULL …. a more graceful error

scenario.

Allocation too big …

not enough free

memory

Automatic Explicit

Access Fast Slower

Fragmentation No Yes

Memory Fragmentation

• Memory fragmentation: the memory allocated on the heap is spread out of the memory space, rather than being concentrated in a certain address space.

Memory FragmentationCode

Negative aspects of fragmentation?(1) can’t make big allocations

(2) losing cache coherency

Fragmentation and Big AllocationsCode

Even if there is lots of memory available, the memory manager can only accept your request if there is a

big enough contiguous chunk.

Automatic Explicit

Access Fast Slower

Fragmentation No Yes

Memory Errors

• Array bounds read

• Array bounds write

Memory Errors

• Free memory read / free memory write

When does this happen in real-world scenarios?

Memory Errors

• Freeing unallocated memory

Vocabulary: “dangling pointer”: pointer that points to memory that has already been freed.

Memory Errors

• Freeing non-heap memory

Memory Errors

• NULL pointer read / write

• NULL is never a valid location to read from or write to, and accessing them results in a “segmentation fault”– …. remember those memory segments?

Memory Errors

• Unitialized memory read

hank childs, university of oregon

object code file file1

machine code

c gcc flag c

input file

configuration file

library libnamemakefilesthere

static library

omultifile development

Documents

hank childs, university of oregon april 3rd, 2015 cis 330: _...

hank childs, university of oregon lecture #6 cis 410/510:...

hank childs, university of oregon november 22 nd, 2013 force...

lbnl vacet activities hank childs computer systems engineer...

hank childs, university of oregon november 15 th, 2013...

hank childs, university of oregon volume rendering primer /...

hank childs, university of oregon oct. 17 th, 2014 cis...

visit : a tool for visualizing and analyzing very large data...

hank childs, university of oregon lecture #1 cis 410/510:...

hank childs, university of oregon lecture #10 cis 410/510:...

hank childs, university of oregon lecture #4 fields, meshes,...

how petascale visualization will change the rules hank...

hank childs, university of oregon lecture #3 fields, meshes,...

hank childs, university of oregon june 3 rd, 2015 cis 330: _...

hank childs, university of oregon jan. 21st, 2013 cis 610:...

hank childs (lbnl) and charles doutriaux (llnl) september...

hank childs, university of oregon nov. 7th, 2014 cis...

hank childs, university of oregon volume rendering, part 3

hank childs, university of oregon volume rendering, pt 1

hank childs, university of oregon may 13th, 2015 cis 330: _...